Distributed Data Reconciliation And Bias Estimation With Non-Gaussian Noise For Sensor Network

Abstract

This thesis considers both Data Reconciliation &40;DR&41; and Bias Estimation &40;BE&41; with non-Gaussian noise in a distributed sensor network environment&46; The distributed DR &40;DDR&41; and distributed BE &40;DBE&41; are derived&44; and the implementation algorithms are developed&46; DDR and DBE are robust to node failures&46; Illustrative examples and application case studies of an experimental-scale chemical plant are presented to demonstrate the proposed DDR and DBE&46; The performance of the Generalized T &40;GT&41;&44; inter-quartile range test cum least-square &40;IQR&43;LS&41; and least-square &40;LS&41; bias estimators used in DBE are analyzed through both theoretical tools and experiments&44; for cases where data are contaminated with outliers&46; The results show that GT bias estimator is the most efficient when outliers are close to good data&46; Furthermore&44; as the theoretical tools relate the estimator type and sample size with the estimation variance&44; it allows one to design an estimator to achieve a specified variance&46;